dissenting.
I agree with much that Justice Scalia says in his dissent, but I write separately because' in certain respects his approach differs from mine. As Justice Scalia ably demonstrates, the Board’s analysis in this case cannot be reconciled with its decisions in cases such as Service Electric Co., 281 N. L. R. B. 633 (1986), and Leveld Wholesale, Inc., 218 N. L. R. B. 1344 (1975). Those decisions rest upon the premises that a striking union inevitably will tend to favor its own members at the expense of replacement workers; that the union cannot reasonably be expected to give balanced representation to the two groups; and that the replacements “can reasonably foresee that, if the union is successful, the strikers will return to work and the strike replacements will be out of a job.” Id., at 1350. Those premises are, to say the least, in considerable tension with the Board’s refusal to presume, without direct evidence of employee preferences, that an employer may in good faith doubt that replacement workers support the striking union.
Justice Scalia’s dissent, as I read it, rests upon the belief that the Board was correct in Service Electric and Leveld Wholesale, and that its decision in the instant case is therefore substantively irrational. Certainly the views expressed in Service Electric and Leveld Wholesale accord with my own understanding of industrial reality. It seems to me eminently foreseeable that a striking union will disfavor the workers who have been hired to break the strike; that the union will attempt, as an element of the ultimate settlement, to secure the discharge of replacement employees; and that the replacements will be aware of the antagonism between the union’s interests and their own.1 But if the expert agency were to determine that the participants in the *799collective-bargaining process no longer behave in this fashion, and if it consistently acted upon this determination, I cannot say at this juncture that the Board’s decision would be irrational. To invalidate the Board’s order in the present case, it is not necessary to assert that the decision is based upon an implausible assessment of industrial reality. Rather, it is enough to say that the Board in this case has departed, without explanation, from principles announced and reaffirmed in its prior decisions. The agency has made no effort to explain the apparent inconsistency between the decision here and its analyses in Service Electric and Leveld Wholesale, and its order is invalid on that basis alone.2
1 am struck, moreover, by the Board’s lack of empirical support for its position — a significant point in view of the fact that for 25 years the Board presumed that replacement workers opposed the striking union. If the Board’s refusal to adopt such a presumption is based, at least in part, on policy concerns ie. g., the fear that employers would abuse the bargaining process by “hiring their way out” of their statutory duty), it seems reasonable to expect the Board to show (or at least to assert) that such abuses actually occurred during the period the presumption was in place. I am also troubled by the fact, noted in The Chief Justice’s concurring opinion, ante, at 797, that while the Board appears to require that good-faith doubt be established by express avowals of individual employees, other Board policies make it practically impossible for the employer to amass direct evidence of its workers’ views.3 The point, I emphasize, is that the propri*800ety of the no-presumption rule cannot be determined simply by asking whether the rule, in isolation, is irrational or rests on a demonstrably misguided view of the facts. Rather, the reviewing court also must ask whether the agency’s decision is the product of an adequate deliberative process and is consonant with other agency pronouncements in analogous areas.
Perhaps the difference between my approach and that of Justice Scalia is one only of emphasis, but I think that the difference is worth noting. Rarely will a court feel so certain of the wrongness of an agency’s empirical judgment that it will be justified in substituting its own view of the facts. But courts can and should review agency decisionmaking closely to ensure that an agency has adequately explained the bases for its conclusions, that the various components of its policy form an internally consistent whole, and that any apparent contradictions are acknowledged and addressed. This emphasis upon the decisionmaking process allows the reviewing court to exercise meaningful control over unelected officials without second-guessing the sort of expert judgments that a court may be ill equipped to make. Such an approach also affords the agency a broad range of discretion. Confronted with a court’s conclusion that two of its policy pronouncements are inconsistent, the agency may choose for itself which path to follow, or it may attempt to explain why no contradiction actually exists.
This Court has never held that the Board is required by statute to recognize the good-faith doubt defense, and the Board’s power to eliminate that defense remains an open *801question. The Board has not purported to take that step, however, and the agency has articulated no legitimate basis for its conclusion that the employer in this case lacked a good-faith doubt as to the union’s majority support. The Board may not assert in one line of cases that the interests of a striking union and replacement workers are irreconcilably in conflict, and proclaim in a different line of decisions that no meaningful generalizations can be made about the union sentiments of the replacement employees. I therefore conclude that the judgment of the Court of Appeals should be affirmed.
I respectfully dissent.
Justice Scalia, with whom Justice O’Connor and Justice Kennedy join, dissenting.The Court makes heavy weather out of what is, under well-established principles of administrative law, a straightforward case. The National Labor Relations Board (NLRB or Board) has established as one of the central factual determinations to be made in § 8(a)(5) unfair-labor-practice adjudications, whether the employer had a reasonable, good-faith doubt concerning the majority status of the union at the time it requested to bargain. The Board held in the present case that such a doubt was not established by a record showing that at the time of the union’s request a majority of the bargaining unit were strike replacements, and containing no affirmative evidence that any of those replacements supported the union. The question presented is whether that factual finding is supported by substantial evidence. Since the principal employment-related interest of strike replacements (to retain their jobs) is almost invariably opposed to the principal interest of the striking union (to replace them with its striking members) it seems to me impossible to conclude on this record that the employer did not have a reasonable, good-faith doubt regarding the union’s majority status. The Board’s factual finding being unsupported by substantial evidence, it cannot stand. I therefore dissent from the judg*802ment reversing the Fifth Circuit’s refusal to enforce the Board’s order.
I
As the Court describes, the union was certified in 1970. In 1979, before the strike, the bargaining unit was composed of 27 employees. After the strike began, 5 employees crossed the picket line and 29 new employees were hired to replace the strikers. On July 16, the union offered to return to work under the terms of respondent’s last prestrike proposal. On July 20, respondent rejected this offer, withdrew recognition from the union, and refused to bargain further, stating that it doubted that the union represented a majority of the employees in the bargaining unit. On that date, according to the Board, the bargaining unit consisted of 49 employees: 19 strikers, 5 employees who had crossed the picket line to return to work, and 25 strike replacements. The union filed an unfair-labor-practice charge with the Board, and the Board’s General Counsel filed a complaint charging that respondent had violated § 8(a)(5) (and thereby § 8(a)(1)) of the National Labor Relations Act (NLRA), 49 Stat. 452, as amended, 29 U. S. C. §§ 158(a)(1) and (5). After a formal hearing, the Administrative Law Judge (ALJ) found that respondent, at the time it withdrew recognition, had an objectively reasonable, good-faith doubt that the union represented a majority of the employees in the bargaining unit. The General Counsel introduced no evidence that the union in fact commanded the support of a majority of the employees, and the ALJ dismissed the complaint.
The Board reversed and entered an order finding that respondent had violated the NLRA. The Board began by reciting its longstanding rule that “ ‘[a]n employer who wishes to withdraw recognition after a year may do so ... by presenting evidence of a sufficient objective basis for a reasonable doubt of the union’s majority status at the time the employer refused to bargain.’” 287 N. L. R. B. 350, 352 (1987) (quoting Station KKHI, 284 N. L. R. B. 1339, 1340 (1987)). *803Purporting to evaluate respondent’s action under this standard, the Board concluded that respondent had not established a reasonable basis for doubting the union’s majority status. First, the Board stated that there was no cause to doubt that the five strikers who crossed the picket line to return to work supported the union, because “[t]he failure of employees to join an economic strike may indicate their economic concerns rather than a lack of support for the union.” 287 N. L. R. B., at 352. Second, relying on its decision in Station KKHI, the Board stated that the fact that 25 employees in the bargaining unit were strike replacements provided “no evidentiary basis for reasonably inferring the union sentiments of the replacement employees as a group.” Id., at 352-353. Third, the Board discounted the statements of 6 employees criticizing the union, because “[e]ven attributing to them the meaning most favorable to the Respondent, it would merely signify that 6 employees of a total of approximately 50 did not desire to keep the Union as the collective-bargaining representative.” Id., at 353 (footnote omitted).
The Fifth Circuit denied enforcement of the Board’s order on the ground that respondent “was justified in doubting that the striker replacements supported the union,” and that the Board’s contrary conclusion was not supported by substantial evidence. 859 F. 2d 362, 365, 367 (1988).
II
An NLRB unfair-labor-practice action is the form of administrative proceeding known as formal adjudication, governedby the Administrative Procedure Act (APA), 5 U. S. C. §§ 556, 557. See 5 U. S. C. § 554(a). In fact, it is even somewhat more judicialized than ordinary formal adjudication, since it is governed in addition by the procedural provisions of the NLRA itself, which provide, inter alia, that the proceeding “shall, so far as practicable, be conducted in accordance with the rules of evidence applicable in the district courts of the United States under the rules of civil procedure for the *804district courts of the United States,” 29 U. S. C. § 160(b). Among the attributes of formal adjudication relevant here, the agency opinion must contain “findings and conclusions, and the reasons or basis therefor, on all the material issues of fact, law, or discretion presented on the record,” 5 U. S. C. § 557(c), and a reviewing court must “hold unlawful and set aside agency action, findings, and conclusions found to be . . . unsupported by substantial evidence,” 5 U. S. C. §706(2)(E); accord, 29 U. S. C. § 160(f) (“[T]he findings of the Board with respect to questions of fact if supported by substantial evidence on the record considered as a whole shall ... be conclusive”).
The Board’s factual finding challenged in the present case is that there was no “‘sufficient objective basis for a reasonable doubt of the union’s majority status at the time of the employer refused to bargain.’” 287 N. L. R. B., at 352(quoting Station KKHI, 284 N. L. R. B., at 1340). The Board has held for many years that an employer’s reasonable, good-faith doubt as to a certified union’s continued majority status is a defense to an unfair-labor-practice charge for refusal to bargain, in the sense that it shifts to the General Counsel the burden to “come forward with evidence that on the refusal-to-bargain date the union in fact did represent a majority of employees in the appropriate unit.” Stoner Rubber Co., 123 N. L. R. B. 1440, 1445 (1959) (emphasis in original). The leading case on the subject, cited approvingly in the Station KKHI opinion that formed the basis for the Board’s holding here, described the good-faith doubt defense as follows:
“We believe that the answer to the question whether the Respondent violated Section 8(a)(5) of the Act . . . depends, not on whether there was sufficient evidence to rebut the presumption of the Union’s continuing majority status or to demonstrate that the Union in fact did not represent the majority of the employees, but upon whether the Employer in good faith believed that *805the Union no longer represented the majority of the employees. . . .
By its very nature, the issue of whether an employer has questioned a union’s majority in good faith cannot be resolved by resort to any simple formula. It can only be answered in the light of the totality of all the circumstances involved in a particular case. But among such circumstances, two factors would seem to be essential prerequisites to any finding that the employer raised the majority issue in good faith in cases in which a union had been certified. There must, first of all, have been some reasonable grounds for believing that the union had lost its majority status since its certification. And, secondly, the majority issue must not have been raised by the employer in a context of illegal antiunion activities . . . .” Celanese Corp. of America, 95 N. L. R. B. 664, 671-673 (1951).
The Board purported to be proceeding on the basis that Celanese was still the law, and thus that “the totality of all [sic] the circumstances” in the present case did not establish reasonable grounds for doubting majority status. The precise question presented is whether there was substantial evidence to support this factual finding. There plainly was not.
As described above, of the 49 employees in the bargaining unit at the time of respondent’s refusal to bargain, a majority (25) were strike replacements, and another 5 were former employees who had crossed the union’s picket line. It may well be doubtful whether the latter group could be thought to support the union, but it suffices to focus upon the 25 strike replacements, who must be thought to oppose the union if the Board’s own policies are to be believed. There was a deep and inherent conflict between the interests of these employees and the interests of the union. As the Board’s cases have explained:
*806“Strike replacements can reasonably foresee that, if the union is successful, the strikers will return to work and the strike replacements will be out of a job. It is understandable that unions do not look with favor on persons who cross their picket lines and perform the work of strikers.” Leveld Wholesale, Inc., 218 N. L. R. B. 1344, 1350 (1975).
“The Union had been bargaining agent for those discharged employees and there can be no question that the Union’s loyalty lay with these employees. The interests of the discharged employees were diametrically opposed to those of the strike replacements. If the discharged employees returned to work, the strike replacements would lose their jobs.” Beacon Upholstery Co., 226 N. L. R. B. 1360, 1368 (1976) (footnote omitted).
The Board relies upon this reality of “diametrically opposed” interests as the basis for two of its rules: First, that an employer does not commit an unfair labor practice by refusing to negotiate with the incumbent union regarding the terms and conditions of the replacements’ employment. See Service Electric Co., 281 N. L. R. B. 633, 641 (1986). Second, that the union’s duty of fair representation does not require it to negotiate in the best interests of the strike replacements regarding the terms and conditions of their employment — in other words, that the union may propose “negotiations leading to replacements being terminated to make way for returning strikers,” ibid. See id., at 639 (“[I]t is not logical to expect [the striking bargaining] representative to negotiate in the best interests of strike replacements during the pendency of a strike”) (internal quotation omitted; citation omitted); id., at 641 (even after the strike has ended, the “inherent conflict between the two groups remains” until the underlying contractual dispute has been resolved); Leveld Wholesale, supra, at 1350 (“It would be asking a great deal of any union to require it to negotiate in the best interests of strike replace*807ments during the pendency of a strike, where the strikers are on the picket line”).
The respondent in this case, therefore, had an employee bargaining unit a majority of whose members (1) were not entitled to have their best interests considered by the complainant union, (2) would have been foolish to expect their best interests to be considered by that union, and indeed (3) in light of their status as breakers of that union’s strike, would have been foolish not to' expect their best interests to be subverted by that union wherever possible. There was, moreover, not a shred of affirmative evidence that any strike replacement supported, or had reason to support, the union. On those facts, any reasonable factfinder must conclude that the respondent possessed, not necessarily a certainty, but at least a reasonable, good-faith doubt, that the union did not have majority support. At least three Courts of Appeals have effectively agreed with this assessment, considering strike replacements as opposed to the union in reversing Board findings of no reasonable, good-faith doubt. See Soule Glass & Glazing Co. v. NLRB, 652 F. 2d 1055, 1110 (CA1 1981); National Car Rental System, Inc. v. NLRB, 594 F. 2d 1203, 1206-1207 (CA8 1979); NLRB v. Randle-Eastern Ambulance Service, Inc., 584 F. 2d 720, 728-729 (CA5 1978).
In making its no-reasonable-doubt finding, the Board relied upon its decision in Station KKHI, which stated:
“[T]he hiring of permanent replacements who cross a picket line, in itself, does not support an inference that the replacements repudiate the union as collective-bargaining representative. ... In this regard, an employee may be forced to work for financial reasons, or may disapprove of the strike in question but still desire union representation and would support other union initiatives. The presumption of union disfavor is therefore not factually compelling.” 284 N. L. R. B., at 1344 (footnotes omitted).
*808The Court finds this reasoning persuasive: “Economic concerns, for instance, may force a replacement employee to work for a struck employer even though he otherwise supports the union and wants the benefits of union representation.” Ante, at 789. These responses are entirely inadequate. The question is not whether replacement employees accept employment for economic reasons. Undoubtedly they do — the same economic reasons that would lead them to oppose the union that will likely seek to terminate their employment. Nor is the question whether replacements would like to be represented by a union. Some perhaps would. But what the employer is required to have a good-faith doubt about is majority support, not for “union representation” in the abstract, but for representation by this particular complainant union, at the time the employer withdrew recognition from the union.
Also embarrassingly wide of the mark is the Court’s observation that “[u]nions do not inevitably demand displacement of all strike replacements.” Ante, at 790. It is not necessary to believe that unions inevitably demand displacement of all strike replacements in order to doubt (as any reasonable person must) that strike replacements support a union that is under no obligation to take their employment interests into account, and that is almost certain to demand displacement of as many strike replacements as is necessary to reinstate former employees. The Court does not accurately describe my position, therefore, when it suggests that I seek “to force the Board to apply a presumption based on the premise that unions always make such demands.” Ante, at 793, n. 11. I seek to force the Board, as the APA requires, to give objectively reasonable probative effect to the reality (expressed in the Board’s own opinions and made the basis for rules of law it has adopted) that unions almost always make stlch demands, and that replacement employees know that. That is enough to establish this employer’s good-faith doubt.
*809The Court asserts that “the facts of this case” belie the proposition stated supra, at 808, that a union “is almost certain to demand the displacement of as many strike replacements as is necessary to reinstate former employees,” because the union in this case “did not negotiate for the discharge of replacements as a condition for settling the strike.” Ante, at 791, n. 9. That is not true, and even if it were true it would not be determinative of the issue here. According to the Board, this is what happened:
“On 16 July the Union, on behalf of the striking employees, made an unconditional offer to return to work, thereby ending the strike. Later on the same day, the Union notified the Respondent that the bargaining unit employees had accepted the Respondent’s 25 May collective-bargaining proposal.” 287 N. L. R. B., at 351 (emphasis added).
Surely an offer “on behalf of the striking employees ... to return to work” can only be accepted by allowing the striking employees to return to work. Does the Court really mean to interpret the union’s action as agreement that the strike replacements shall stay on the job under the terms of the May 25 collective-bargaining proposal, and the strikers remain unemployed? Or as a proposal that the employer should double its work force, paying both the replacement workers and the returning strikers under the terms of the May 25 offer? Surely the very most that can be said is that the union’s offer left the status of the replacement workers for later negotiation. No more is necessary to establish a reasonable doubt that the replacement workers would support the union — which, in any such negotiations, could be expected (indeed, would have a legal obligation) to seek displacement of the strikebreakers by the returning strikers. As the Board said in Service Electric:
“[E]ven if the strike be deemed to have ended by virtue of the Union’s announcement that it had been termi*810nated, there is no basis for concluding that the Union suddenly is better able ‘to negotiate in the best interests of strike replacements,’ Leveld Wholesale, . . . 218 NLRB 1344, than it previously had been able to do. The inherent conflict between the two groups remains.” 281 N. L. R. B., at 641.
The Court mentions only as an afterthought the fundamental conflict of interests that is at the center of this case:
“Moreover, even if the interests of strikers and replacements conflict during the strike, those interests may converge after the strike, once job rights have been resolved. Thus, while the strike continues, a replacement worker whose job appears relatively secure might well want the union to continue to represent the unit regardless of the union’s bargaining posture during the strike.” Ante, at 792 (emphasis in original).
The trouble with this is that it posits a species of replacement worker that will rarely exist unless and until the union has agreed (as it had not in this case) to accept the replacements’ employment status — i. e., until “job rights have been resolved.” How can there be “a replacement worker whose job appears relatively secure” when the employer agrees to negotiate in good faith with a union that will surely seek the reinstatement of all its strikers? Even a replacement worker who has clear seniority over other replacement workers, and who somehow knows (by what means I cannot imagine) that some of the striking workers no longer want their jobs back,1 has no means of assurance that the union will do *811him the favor of bargaining for the employer to honor his seniority among strikebreakers. It seems overwhelmingly likely that the union will want its returning members to have their old jobs back, or better jobs, regardless of the relative seniority of the strikebreakers who would thereby be displaced. I do not dispute that “replacement workers are capable of looking past the strike in considering whether or not they desire representation by the union,” ibid.— in the same way that a man who is offered $1 million to jump off a cliff is capable of looking past the probable consequence of his performance to contemplate how much fun he would have with $1 million if he should survive. Surely the benefits strike replacements anticipate from their poststrike representation by this particular union must be expected to weigh much less heavily in their calculus than the reality that if this particular union does the bargaining and gets its way, they will not have poststrike jobs.
The Court’s only response to this is that the union’s ability to achieve displacement of the strike replacements will depend upon its bargaining power. Its bargaining power could conceivably be so weak, and a strike replacement might conceivably so prefer this union over other alternatives, that he would be willing to take the chance that the union will try to oust him. Ante, at 790-791. I suppose so. It might also be that one of the strike replacements hopes the union will continue as the bargaining representative because, as the employer knows, the union president is his son-in-law. The Board Counsel is entirely free to introduce such special circumstances. But unless they appear in the record, the reasonableness of the employer’s doubt must be determined on the basis of how a reasonable person would assess the probabilities — and it is overwhelmingly improbable that a *812strikebreaking replacement so much prefers the incumbent union to some other union, or to no union at all, that he will bet his job the union is not strong enough to replace him. The wager is particularly bad because it is so unnecessary, since he and his fellow replacements could achieve the same objective, without risking their jobs in the least, by simply voting for that union, after the strike is over, in a new certification election.
I reiterate that the burden upon the employer here was not to demonstrate 100% assurance that a majority of the bargaining unit did not support the union, but merely “reasonable doubt” that they did so. It seems to me absurd to deny that it sustained that burden.
Ill
The Court never directly addresses the question whether there was substantial evidence to support the Board’s conclusion that respondent had not established a reasonable good-faith doubt of the union’s majority status. Indeed, it asserts that that question is not even at issue, since “[t]he question on which we granted the Board’s petition for certiorari is whether, in assessing whether a particular employer possessed a good-faith doubt, the Board must adopt a general presumption of replacement opposition to the union.” Ante, at 778, n. 2. That is the equivalent of characterizing the appeal of a criminal conviction, in which the defendant asserts that the indictment should have been dismissed because all the evidence demonstrated that he was not at the scene of the crime, as involving, not the adequacy of the evidence, but rather the question whether the jury was required to adopt the general presumption that a person cannot be in two places at the same time. No more in administrative law than in criminal law is the underlying question altered by characterizing factual probabilities as presumptions. The two are one and the same. The only reason respondent asserts, and the Fifth Circuit held, that the Board had to adopt the “pre*813sumption” of replacement opposition to the union (I place the word in quotation marks because, as I shall describe, that terminology is misleading) is that to refuse to apply that “presumption” for the reason the Board gave (viz., that it is not in fact an accurate assessment of probabilities) is to deny evidence its inherently probative effect, and thus to produce a decision that is not supported by substantial evidence. The Board’s framing of the question presented, like its opinion in this case, invites us to confuse factfinding with policymaking. The Court should not so readily have accepted the invitation.
Prior to its decision in Station KKHI, the Board well understood the inevitable logic set forth in Part II above, and had held that an employer has an objectively reasonable basis for doubting the union’s majority status when the majority of employees in the bargaining unit are permanent replacements, and there is no other indication regarding the replacements’ sentiments toward the union. See Beacon Upholstery Co., 226 N. L. R. B., at 1368; Titan Metal Mfg. Co., 135 N. L. R. B. 196, 215 (1962); Stoner Rubber Co., 123 N. L. R. B., at 1445. In a case called Cutten Supermarket, 220 N. L. R. B. 507 (1975), the Board departed, in dictum, from this well-established precedent. There, as the Board describes the case in Station KKHI, 284 N. L. R. B., at 1341, the Board “inexplicably stated,” with respect to strike replacements: “[I]t is a well-settled principle that new employees are presumed to support the union in the same ratio as those whom they have replaced.” 220 N. L. R. B., at 509. This dictum became a Board holding in Pennco, Inc., 250 N. L. R. B. 716 (1980), where the Board stated: “The Board has long held that [the presumption of strike replacement support for the union] applies as a matter of law, and it is incumbent upon Respondent to rebut it even, and perhaps especially, in the event of a strike.” Id., at 717 (emphasis added). As the Board acknowledged in Station KKHI:
*814“The Board in Pennco cited no cases in support of its assertion that the presumption in question was ‘long held.’ Indeed, . . . this presumption was not ‘long held’ at all, but in fact was not articulated in any fashion until Cuíten Supermarket in 1975, only 5 years prior to Pennco, and even then (i. e., in Cuíten) in dictum and without supporting rationale or precedent.” 284 N. L. R. B., at 1343.
Unsurprisingly, given the feeble support for this presumption, the Courts of Appeals (I am still repeating the account in Station KKHT) “uniformly rejected it.” Ibid.
In Station KKHI, which is the case that established the framework of reasoning for the decision we review today, the principal issue before the Board was the Pennco presumption of replacement support for the union. See 284 N. L. R. B., at 1340. The Board abandoned it. See id., at 1344. The Board went further, however, and here is the error that infects the present opinion:
“On the other hand, we find the contrary presumption equally unsupportable. Thus, the hiring of permanent replacements who cross a picket line, in itself, does not support an inference that the replacements repudiate the union as collective-bargaining representative.” Ibid.
The mistake here is to treat as equivalent elements of decisionmaking, the presumption that strike replacements do support the union, and the evidentiary inference that strike replacements do not support the union. They are not different applications of the same device, and it does not display a commitment to be governed only by the “real facts” to reject both the one and the other. The former was applied “as a matter of law,” Pennco, supra, at 717, and not as the product of inference, which is “[a] process of reasoning by which a fact or proposition sought to be established is deduced as a logical consequence from other facts, or a state of facts, already proved or admitted.” Black’s Law Dictionary 700 (5th ed. *8151979). One can refer to the product of an inference as a presumption: If one knows the identity of the sole Englishman in a certain remote part of Africa, and encounters there a white man in pith helmet sipping a gin and tonic, it is perfectly appropriate to say “Dr. Livingston, I presume.” But that sort of presumption, which the text writers used to call “presumption of fact,” see 9 J. Wigmore, Evidence §2491, p. 304 (J. Chadbourn rev. ed. 1981), is quite different from the Pennco-type “presumption of law” insofar as concerns both agency power and judicial review, as I shall proceed to explain.
It is the proper business of the Board, as of most agencies, to deal in both presumptions (i. e., presumptions of law) and inferences (presumptions of fact). The former it may create and apply in the teeth of the facts, as means of implementing authorized law or policy in the course of adjudication. An example is the virtually irrebuttable presumption of majority support for the union during the year following the union’s certification by the Board, Station KKHI, 284 N. L. R. B., at 1340. The latter, however — inferences (or presumptions of fact) — are not creatures of the Board but its masters, representing the dictates of reason and logic that must be applied in making adjudicatory factual determinations. Whenever an agency’s action is reversed in court for lack of “substantial evidence,” the reason is that the agency has ignored inferences that reasonably must be drawn, or has drawn inferences that reasonably cannot be. As I have discussed above, that is what happened here.2
*816Of course the Board may choose to implement authorized law or policy in adjudication by forbidding a rational inference, just as it may do so by requiring a nonrational one (which is what a presumption of law is). And perhaps it could lawfully have reached the outcome it did here in that fashion — saying that even though it must reasonably be inferred that an employer has good-faith doubt of majority status when more than half of the bargaining unit are strike replacements whose job rights have not been resolved, we will not permit that inference to be made. (This would produce an effect close to a rule of law eliminating the good-faith doubt defense except for cases in which the employer can demonstrate, by employee statements, lack of support for the union.) But that is not what the agency did here. It relied on the reasoning of Station KKHI, which rested upon the conclusion that, as a matter of logic and reasoning, “the hiring of permanent replacements who cross a picket line, in itself, does not support an inference that the replacements repudiate the union as collective-bargaining representative.” Id., at 1344. That is simply false. It is bad factfinding, and must be reversed under the “substantial evidence” test.
It is true that Station KKHI added, seemingly as a makeweight: “Moreover, adoption of this presumption would disrupt the balance of competing economic weapons long established in strike situations and substantially impair the employees’ right to strike by adding to the risk of replacement the risk of loss of the bargaining representative as soon as replacements equal in number to the strikers are willing to cross the picket line.” Ibid. There are several reasons why *817we cannot allow this seeming policy justification to suffice for sustaining the agency’s action.
First of all, it is set forth as a reason for not adopting a counterfactual presumption, rather than (what would have been necessary to validate the agency’s action) a reason for ignoring legitimate factual inferences. It is one thing to say: “The facts do not support conclusion X, and we decline to impose conclusion X as a matter of law, since that would have adverse policy consequences.” It is quite another thing to say: “Even though the facts require conclusion X, we reject it for policy reasons.” The former is what the Board has said here, and the latter is what it would have to say to support its decision properly on policy grounds. The agency has set forth a reason for rejecting the suggestion that it ignore the facts; having found that, on the record, the facts were the opposite of what the agency believed (j. e., there was reasonable good-faith doubt), we have no idea whether the agency would regard the same reason as adequate basis for now accepting a suggestion that it ignore the facts. Under long-established principles of judicial review, we cannot make that yet-to-be-made decision on the agency’s behalf, but must remand so that the Board may do so. SEC v. Chenery Corp., 318 U. S. 80, 88 (1943).
Second, by upholding as a counterfactual policy determination a ruling that was made, and defended before the Fifth Circuit, as ordinary factfinding plus the refusal to adopt a counterfactual policy determination, we would be depriving respondent of possible legal defenses that it had no occasion to present to the courts. Section 8(a)(5) of the NLRA is violated only if the employer refuses to bargain “with the representatives” of his employees. 29 U. S. C. § 158(a)(5). The Act does not define the term “representatives,” except to say that it “includes any individual or labor organization.” § 152 (4). Specifically, it does not say or anywhere suggest that a certified union is the “representative” for purposes of § 158(a) (5) unless and until it is decertified. Because the Board has *818long acknowledged the good-faith doubt defense, there has been no occasion to test in the courts the proposition that the employer can be liable for refusing to bargain with a certified union that patently does not have majority support. The only presumption of law that is applied to effect a policy exception to this defense — the almost irrebuttable presumption of union support for one year after its certification — is arguably authorized as an implementation of the policy of § 159(e) (2), which precludes certification elections more frequently than annually. Even if, moreover, the Board can generally require the employer to bargain with a union that is not a “representative” in the sense of having majority support, there is the further question whether it can require him to bargain with a union that is a “representative” neither in that sense nor in the sense that it is obligated to bargain in the best interests of the majority of employees. See my earlier discussion of the Board’s rule that the union has no duty to represent the interest of replacement employees, supra, at 806-807. The Board did not have to confront these issues in the present case, because it did not purport to be deciding the case on the assumption that the union lacked majority support. It found respondent guilty of an unfair labor practice on the ground that there was, in fact, no reasonable doubt of the union’s majority status; and it is exclusively that finding which respondent challenged, both here and in the Fifth Circuit. If we permitted the Board’s order to be enforced on the quite different ground that it does not matter whether respondent had a reasonable, good-faith doubt, we not only would be making for the Board a decision it has not yet reached, but also would be depriving respondent of judicial review of that decision.3
*819* * *
Despite the fact that the NLRB has explicit rulemaking authority, see § 156, it has chosen — unlike any other major agency of the Federal Government — to make almost all its policy through adjudication. It is entitled to do that, see NLRB v. Bell Aerospace Co., 416 U. S. 267, 294-295 (1974), but it is not entitled to disguise policymaking as factfinding, and thereby to escape the legal and political limitations to which policymaking is subject. Thus, when the Board purports to find no good-faith doubt because the facts do not establish it, the question for review is whether there is substantial evidence to support that determination. Here there is not, and the Board’s order should not be enforced.
What the Court has permitted the Board to accomplish in this case recalls Chief Justice Hughes’ description of the unscrupulous administrator’s prayer: “ ‘Let me find the facts for the people of my country, and I care little who lays down the general principles.’” Address before Federal Bar Association, February 12, 1931, quoted by Frank, J., in United States v. Forness, 125 F. 2d 928, 942 (CA2 1942), reprinted in 13 The Owl of Sigma Nu Phi 9, 12 (Feb. 1931). I respectfully dissent.
See Belknap, Inc. v. Hale, 463 U. S. 491, 513-514 (1983) (opinion concurring in judgment) (“During settlement negotiations, the union can be counted on to demand reinstatement for returning strikers as a condition for any settlement”).
See Greater Boston Television Corp. v. FCC, 143 U. S. App. D. C. 383, 394, 444 F. 2d 841, 852 (1970) (“[A]n agency changing its course must supply a reasoned analysis indicating that prior policies and standards are being deliberately changed, not casually ignored, and if an agency glosses over or swerves from prior precedents without discussion it may cross the line from the tolerably terse to the intolerably mute”) (footnote omitted), cert. denied, 403 U. S. 923 (1971).
The NLRB has recently reaffirmed its rule that an employer must meet the same good-faith doubt standard in order to poll its employees, petition *800the Board for an election, or withdraw recognition from the union. Texas Petrochemicals Corp., 296 N. L. R. B. 1057, 1064 (1989). If good-faith doubt can be established only by the express statements of individual workers, the employer is placed in a difficult bind. See Mingtree Restaurant, Inc. v. NLRB, 736 F. 2d 1295, 1297 (CA9 1984) (“By the Board’s reasoning, an employer in doubt of the union’s majority status would be allowed to take a poll only when it had no actual need to do so, that is, when it already had sufficient objective evidence to justify withdrawal of recognition”).
The Court emphasizes that “at the time of respondent’s withdrawal of recognition from the Union, there were only 19 strikers and 25 replacements. ” Ante, at 792, n. 9. It is easy for the Court to know this, following a full-dress formal adjudication inquiring into the historical fact. One wonders how or why the replacement workers could be thought to have known it at the time. Surely the union did not advertise the fact that its strike force was dwindling, and it is unlikely that the strikebreakers learned it from the remaining strikers in the course of routine frater*811nization. Perhaps they knew, but they most probably did not; and whether the employer’s doubt about replacement support for the union was reasonable depends, of course, upon whether he reasonably assessed the probabilities.
The Court apparently believes that it is all right for the Board to ignore proper factfinding now, so long as it promises to make proper factfinding later. The Court says that “assuming for the sake of argument that . . . unions ‘almost certain[ly]’ demand displacement of all strike replacements . . . , such demands will be a factor in the Board’s analysis in most cases. There is no reason, however, to force the Board to apply a presumption based on the premise that unions always make such demands . . . .” Ante, at 793, n. 11. This presumably means that when the respondent, having been compelled by the Board to negotiate with this *816union, is presented with a proposal for displacement of strike replacements, it may then break off negotiations on the grounds that it has a good-faith doubt of the union’s majority status. I doubt very much that this is what the Board has in mind. But even if it is, it does not justify the Board’s compelling the respondent to negotiate in the first place, when any reasonable person must have entertained a “reasonable doubt” of the union’s majority status — which the Board continues to affirm is enough.
The Court says that it “need not determine whether the Board’s policy considerations alone would justify its refusal to adopt the presumption urged by respondent, because we find the Board’s decision not irrational as a factual matter.” Ante, at 795-796, n. 13. For the reasons just discussed in text, the Board’s policy considerations alone could not possibly be *819held to have justified the refusal in the present case. The Court’s statement does make perfectly clear, however, that today’s judgment ultimately rests upon the determination that it was “not irrational as a factual matter” for the Board to deny on the present record that respondent had a reasonable, good-faith doubt. That determination is simply not credible.